Study On Automatic Dosing System Of Activated Carbon In Urban Sewage Plant Based On BP Neural Network

In recent years,powdered activated carbon(PAC)with the advantages of efficient adsorption and low cost,is no longer used to deal with emergent pollution incidents.Actually,it has developed into a mainstream technology of the advanced treatment.However,the PAC dosing system is still in the artificial control stage,which means the technical staffs are required to determine the dosage of PAC based on experience.As a result,the effluent quality is instable,and the consumption of PAC is great.Therefore,how to adjust the dosage of PAC with the influent quality automatically is becoming an urgent issue,which should be solved when the PAC dosing system is in industrial application.It is of great significance for the small and medium-sized urban sewage treatment plants to meet discharge standards and reduce energy comsumption.This paper was made after the advanced treatment process of an urban sewage treatment plant in Jiashan County of Zhejiang Province.The PAC dosing system was studied and its main influences were determined as raw water COD,pH and flow rate.In this paper,in order to solve issues such as large delay,nonlinear and complexity of PAC dosing system,the BP neural network used in intelligent control was proposed as the feedforward predictive dosing controller,meanwhile the BP neural network feedforward prediction and PID feedback were established as a composite control system program.The main results were as follows:1.According to the 46 results of jar test,BP neural network and multiple linear regression were used to build PAC dosing system feedforward control model.The three-layer structure was adopted in the BP neural network model,in which the input layer node was 2,the output layer node was 1 and the hidden layer node was 11.The unexamined sample was simulated by the offline training model,conforming the fitting degree R2=0.968,the root mean square error RMSE=0.0091.The expression of the multivariate linear regression model was U=0.0170X1+0.0020X2-0.2078.After fitting in the same way,it was confirmed that the fitting degree R2=0.909,and RMSE=0.0145.Compared with the simulation results,it could be verified that the feedforward controller of the PAC dosing system with BP neural network model was superior than another,which could also adapt to different water quality changes with high prediction accuracy and great learning ability.2.The transfer function expression was obtained by theoretical analysis of the model of the controlled object.The soaking curve was obtained according to the soared experiment of the PAC dosing,the inflection point of the step response curve was found as(50,49.78),and the transfer function was confirmed as:G0(s)=5.4/(1+33.75s)(1+16.875s)e-26.6s.The three parameters of the PID controller weredetermined by the critical scale method:KP=0.36;KI=0.006;KD=5.4.The step response curve of the system showed that the excess COD of effluent was more than 40%,and the time required for stabilization was about 300 min.3.In the Simulink environment of Matlab,the feedforward BP neural network predictive-feedback PID compound control system was simulated.The results showed that the overshoot of COD in the composite control system was less than 20%and the adjustment time was about 70min.The PC with WinCC 7.0 was used as the host computer in the demonstration project to carry on the data transmission through the ethernet and the PLC station.The communication between the PC and the PLC was connected by OPC communication tool.According to the formula q=x/10?Q,the PAC was fed with the metering pump flow.After a one-month trial period with automatic control model,the effluent COD compliance rate rose to 90.63%,which was 8.88%higher than the artificial control model.Moreover the daily consumption of PAC was reduced by 16.61%,and the monthly use of PAC was reduced by 135,000 yuan.